Zipf's law for cities is probably the most famous regularity in social sciences. So much that, a hundred years of publication later, its status is not clear: is it a law of social organisation? Is it an instrument of description of city size distributions? Is it an element of validation of geographical objects (cities and systems of cities in particular)? Empirical estimations of the rank-size parameters are very numerous and contradict each other. In this study, we present the results of a reproducible meta-analysis of the largest pool of papers regarding this issue, obtained from the collection of data made open and the construction of an online interactive application which allows the reader to explore this literature. We find that a large part of the variations observed in the measure of Zipf's coefficient is unnecessary as it comes from the choice of different technical specifications in the way cities are defined, whereas some of the current theories to explain the remaining share of variations are challenged by our results.